Short-term Load Forecasting Based On LSTM Deep Neural Network And Improved Kernel Extreme Learning Machine

Power system load forecasting is an important part of power system planning and an important part of ensuring the safety and stability of social production.At present,with the continuous development of technology and the continuous improvement of people's quality of life,power companies are increasingly pursuing more accurate power load forecasting.In the past,many load forecasting research articles did not statistically analyze the influencing factors,but selected some common load influencing factors,such as temperature and meteorology,for load forecasting.With the development of society,people's life content has gradually enriched,and the factors affecting modern power load have gradually changed.This inevitably creates some difficulties for load forecasting and increases the difficulty of forecasting.If we continue to select the general influencing factors for power load forecasting,it will be difficult to meet the modern people's requirements for accuracy.Therefore,this requires people to analyze the power load,clarify the relationship between its internal and external factors,and then carry out load forecasting.At the same time,the single statistical forecasting model in the past is relatively simple,and the prediction accuracy is low.The application of the combined forecasting model in the field of power load forecasting is one of the main hotspots of current research.This paper starts from two aspects,one is the analysis of internal and external influencing factors of short-term electric load forecasting,and the other is the research on short-term electric load forecasting methods.The internal influencing factors of power load refer to the historical load sequence that affects the predicted daily load.The external factors affecting the power load are external factors such as meteorology,temperature and air quality.First,24 power load influencing factors are selected and quantified.Secondly,the Pearson correlation coefficient method is used to analyze the correlation between these 24 influencing factors and the short-term electric load of Yangquan City,to find out the influencing factors that affect the local short-term load,and to analyze the impact of different influencing factors on the local load.Then,the VAR analysis is carried out on the influencing factors of the correlation coefficient,and the influence of the historical data of the external factors with high correlation with Yangquan load on the predicted daily load is studied.The influencing factors of correlation analysis and VAR model selection are relatively comprehensive and accurate,but it is inevitable to increase the input dimension of the prediction model.The excessive input variable dimension will affect the generalization of the prediction model,thus affecting the accuracy of prediction.Finally,principal component analysis can be used to recombine some of the original indicators that are related to each other to replace the original indicators,thus reducing the input index dimension.The data of Yangquan City from March 1 to October 31,2015 was selected for a total of 245 days for empirical analysis.The data of the first 241 days is the training set,and the data of the last 4 days is the prediction set.As one of the research hotspots of traditional neural networks,the Extreme Learning Machine(ELM),this paper combines with the kernel function based on the traditional extreme learning machine,and uses the improved bat optimization algorithm to optimize the nuclear limit learning machine and raise the limit.The stability and accuracy of the learning machine.As one of the hotspots of deep neural network research,long-term and short-term memory neural network(LSTM),the prediction results are compared with the traditional neural network representing BP neural network and adaptive bat algorithm optimization kernel extreme learning machine(IBA-KELM)prediction results.In comparison,the results show the development prospects of deep learning in the field of load forecasting.